Engineering blog
Practical guides, engineering decisions and lessons from working with connected data. Find something useful for what you’re building next.
Engineering & ideas
28 articles · newest first
A dependable import separates valid related records from references that cannot yet be resolved. Stage the data, validate the references and publish parents before children, preserving a report of everything held back.
Read the articleA daily chart should distinguish a measured zero from a day with no complete observation. Start from the reporting calendar and attach evidence of completeness, so missing days do not vanish from the chart.
Read the articleA falling counter deserves review, but it should not erase a missing observation. Classify each reading with its evidence so operators can distinguish a possible reset from a gap in the measurement stream.
Read the articleAn exception inbox should tell an operator what needs attention and why. Keep each finding attached to an order and a responsible role, so a missing receipt does not disappear inside an aggregate total.
Read the articleBuild a small labeled question set, inspect missing and irrelevant results, and test permissions before putting generated answers in front of users.
Read the articleModel machines, service jobs and guides so a maintenance application can explain which evidence belongs to which asset.
Read the articleSeparate messages, summaries and retrieved evidence so an assistant can use relevant context without treating every old message as a current fact.
Read the articleKeep decisions traceable to meetings, index useful passages and return readable evidence instead of disconnected snippets.
Read the articleBuild an import contract, explain rejected rows and verify the accepted batch instead of treating a readable CSV as valid data.
Read the articleConnect customers to purchases, preserve customers with no orders and verify counts before turning the result into a screen.
Read the articleDesign a recoverable indexing workflow for edited documents, interrupted requests and content revisions.
Read the articleDesign a small GenAI workflow that retrieves evidence, checks access, cites the source and declines questions the knowledge base cannot answer.
Read the articleBuild a small retrieval dataset with useful chunks, stable source IDs and consistent embeddings before tuning search.
Read the articleModel the relationship between a machine and its guide, verify the connection, and handle a missing reference before adding an assistant.
Read the articleUse vector search to suggest related support tickets, then check permissions, compare evidence and record a reversible human review decision.
Read the articleBuild a room-booking conflict checker with a four-row SQL example, interval diagrams, boundary tests and a clear plan for concurrent requests.
Read the articleIndex three support notes, compare ranked and all-words search, then update a note without leaving stale search text behind.
Read the articleTurn a small product list into an actionable restock report, with explicit units, predictable results and a simple update workflow.
Read the articleA relational database re-derives relationships with joins at query time; a graph database stores them as structure and walks them. When each one wins.
Read the articleEmbeddings turn meaning into geometry. A plain-language walk through similarity metrics, why brute force dies at scale, HNSW, and what quantization costs.
Read the articleCompare retrieval choices using cost per useful answer, source quality and latency.
Read the articleOriginChainDB's SQL surface now has ROW_NUMBER, RANK, DENSE_RANK, LAG, LEAD and aggregate OVER, plus correlated EXISTS, IN and scalar subqueries.
Read the articleAn implementation note on automatic Idempotency-Key headers, bounded deduplication caches and retry behaviour. Check the SDK version and endpoint contract.
Read the articleOriginChainDB refuses overload with HTTP 429 and a Retry-After computed from the token bucket's refill schedule. The per-API-key limits, and retry code.
Read the articleUse stable operation IDs, stored results and reconciliation to handle retries without assuming external actions are atomic.
Read the articlePer-key TTL lets each record expire on its own: a ttl in seconds on any write, invisible to reads at expiry, compacted in the background. Maximum 365 days.
Read the articleAn OpenAPI spec is what a coding agent reads to generate a client for your API. Ours is vanilla OpenAPI 3.1, one bearer scheme, published at /openapi.json.
Read the articleA guided example of database tools for an AI client. Confirm a compatible server with the team before connecting.
Read the articleKeep exploring
Follow the implementation guides, review current capabilities or bring your workflow to our team.
Build with OriginChainDB
